Comparing approaches to predicting asthma attacks using data

What is it about?

Asthma attacks can be serious and life-threatening, and predicting when they might happen can help patients and their healthcare providers to manage the condition effectively and minimise harm to the patient. Prognostic models, which are tools used to predict when asthma attacks might happen using data about the patient’s symptoms and management, are being widely studied. However, comparing these models can be difficult because they often have big differences in terms of what they consider an asthma attack, how far in the future they predict, who they think the model should be used for, and more. This study aimed to find new predictive models for asthma attacks in adults and to analyse how they are made and how well they work. We looked at 25 studies and found that these models define asthma attacks differently and predict them over various timeframes, from as short as 15 days to as long as 30 months. Most studies used a statistical method called logistic regression to build their models, even though none of the studies that tried multiple methods found it to be the best-performing method. The study also looked to see how different designs of these models affected how well they could predict attacks – both in terms of how often they got it right, and why they got it wrong.  It turns out that sometimes the way these models were built and the data they used meant that they were very good at predicting attacks, but wouldn’t necessarily be the most useful for patients or their healthcare providers.  For example, predicting an asthma attack would happen in the next year with 100% accuracy doesn’t necessarily mean you should straight away start a patient on a new super intensive treatment, especially if it is not convenient for the patient, or has possible side effects.

Featured Image

Why is it important?

The findings suggest that there's a need for input from patients and healthcare providers to improve these models and set rules and guidelines how accurate they need to be, and under what conditions, before they can be used in clinical practice. Working together is our best hope to improve asthma patient outcomes.

Perspectives